LeNet enabled intrusion detection system for iot- networks

Abstract

Now a day's Internet of Thing's (IoT's) is considered growing technology. Ever increasing demand of IoT devices must have reliable channels for communication therefore, security needs to be ensured. Wireless enabled IoT use to have many real-time applications which include smart education, intelligent transportation, smart home, smart grid, health care and farming etc. Data collection can be made possible with the help of massive IoT devices. Intruder directly sends illegal data packets which lead to hijack overall network. IoT-networks have improved human experience in many real-time applications. Cyber-security threats usually limit computational resources. Therefore, detection of cyber-threats is the main problem for IoT-networks. This paper presents a novel concept of leNet model-based IDS to detect cyber- attacks. Although, two benchmark datasets such as IoTID20 and CICIDS2017 are utilized for simulation. LeNet model is compared with other traditional techniques like linear discriminant analysis, quadratic discriminant analysis, support vector machine and multi-layer perceptron. Accordingly, metrics like accuracy, precision, recall, F1-score, confusion matrix and correlation matrix are used for performance evaluation. Extensive analysis is performed which indicates that with IoTIDS20 dataset, LeNet achieves recall-99.91%, precision-99.74%, F1-score-99.83% and accuracy-99.68%. While, using CICIDS2017 dataset, LeNet presents optimal results in terms of recall-97.96%, precision-96.41%, F1-score- 97.18% and accuracy-98.87%. Proposed LeNet model is light weighted and adaptable.